Increasing Risk of Glacial Lake Outburst Floods from Future Third Pole Deglaciation

Articles https://doi.org/10.1038/s41558-021-01028-3

Increasing risk of glacial lake outburst floods from future Third Pole deglaciation

Guoxiong Zheng 1,2,3,13, Simon Keith Allen2,4,13, Anming Bao 1,5 ✉ , Juan Antonio Ballesteros-Cánovas 2,6, Matthias Huss 7,8,9, Guoqing Zhang 10,11, Junli Li 1, Ye Yuan1,3, Liangliang Jiang1,3, Tao Yu 1,3, Wenfeng Chen 3,10 and Markus Stoffel 2,6,12 ✉

Warming on Earth’s Third Pole is leading to rapid loss of ice and the formation and expansion of glacial lakes, posing a severe threat to downstream communities. Here we provide a holistic assessment of past evolution, present state and modelled future change of glacial lakes and related glacial lake outburst flood (GLOF) risk across the Third Pole. We show that the highest GLOF risk is at present centred in the eastern Himalaya, where the current risk level is at least twice that in adjacent regions. In the future, GLOF risk will potentially almost triple as a consequence of further lake development, and additional hotspots will emerge to the west, including within transboundary regions. With apparent increases in GLOF risk already anticipated by the mid-twenty-first century in some regions, the results highlight the urgent need for forward-looking, collaborative, long-term approaches to mitigate future impacts and enhance sustainable development across the Third Pole.

he Hindu Kush–Himalaya, Tibetan Plateau and surrounding floods from moraine-dammed glacial lakes can be triggered by vari- areas are widely known as the Third Pole of the Earth as it ous mechanisms, including intense precipitation and snowmelt26,27, is home to the largest number of glaciers outside the polar and most commonly, from the impact of ice and/or rock avalanches T 1 20,28 regions . Widespread retreat of glaciers is taking place over most into a lake . While robust long-term trends in GLOF frequency of its territory and has accelerated in recent decades as one of the are not evident17,29, the GLOF threat is expected to increase in consequences of global warming2–4. This glacier wasting is associ- response to future warming as lakes expand towards steep and ated with the rapid expansion and new formation of glacial lakes5–8, destabilizing mountain cliffs30,31. With the expansion of commu- bringing both large opportunities and risks9,10. Particularly, when nities, tourism, hydropower, transportation and other crucial sec- water is suddenly released, glacial lake outburst floods (GLOFs)11 tors into exposed areas, substantial risk reduction and adaptation can devastate lives and livelihoods up to hundreds of kilometres strategies will be required to avoid increased impacts on sustain- downstream of their source12,13. This threat is most apparent in the able development in vulnerable mountain regions. Focusing here Third Pole14,15 where the warming rates are distinctly higher than on moraine-dammed proglacial lakes, we draw on a comprehensive the Northern Hemisphere and global mean16, and numerous GLOFs inventory of glacial lakes and past GLOF events across the Third have been recorded, originating from both moraine-dammed and Pole, to model and evaluate how and where GLOF hazard and risk ice-dammed glacial lakes15,17. While outbursts from ice-dammed will change in response to future deglaciation, using a robust and glacial lakes have been concentrated in the Karakoram and Pamir unified evidence-based approach. regions18,19, outbursts from moraine-dammed glacial lakes are observed across the Third Pole, and most frequently along the main Past evolution and present state of glacial lakes Himalayan arc20 where glacial lakes are increasing rapidly in both To understand the present state of glacial lakes across the Third Pole size and number5,6. The impact of a GLOF can extend across inter- and changes over the past decades, all glacial lakes ≥0.01 km2 were national boundaries21, creating severe challenges for early warning mapped based on archival Landsat satellite images from 1990 to and other risk reduction strategies, particularly in critical trans- 2015 (Methods). Our inventory reveals the existence of 26,633 gla- boundary areas22. Despite the severe threat that such large extreme cial lakes in the Third Pole, with a total area of 1,968.8 ± 1.2 km2 and events pose for sustainable mountain development over the Third approximately one-third of them being dammed by moraines (in Pole4, there remains a lack of understanding regarding how and total 7,650 with a combined area of 535.1 ± 0.7 km2; 6,958 of these where GLOF risk will evolve in the future. GLOF assessment over lakes were formed in proglacial areas or are in a transitional state). the Third Pole has typically been focused on moraine-dammed pro- The dataset suggests a clear heterogeneity in their spatial distribu- glacial lakes owing to their potentially large flood volumes23,24, weak tion (Fig. 1 and Supplementary Table 1) as today’s glacial lakes are dam composition25 and clear link to climate change17. Outburst principally distributed along the Hindu Kush–Himalaya–Hengduan

1State Key Laboratory of Desert and Oasis Ecology, Xinjiang Institute of Ecology and Geography, Chinese Academy of Sciences, Urumqi, China. 2Climatic Change Impacts and Risks in the Anthropocene (C-CIA), Institute for Environmental Sciences, University of Geneva, Geneva, Switzerland. 3University of Chinese Academy of Sciences, Beijing, China. 4Department of Geography, University of Zurich, Zurich, Switzerland. 5China-Pakistan Joint Research Center on Earth Sciences, CAS-HEC, Islamabad, Pakistan. 6Dendrolab.ch, Department of Earth Sciences, University of Geneva, Geneva, Switzerland. 7Laboratory of Hydraulics, Hydrology and Glaciology (VAW), ETH Zurich, Zurich, Switzerland. 8Swiss Federal Institute for Forest Snow and Landscape Research (WSL), Birmensdorf, Switzerland. 9Department of Geosciences, University of Fribourg, Fribourg, Switzerland. 10Institute of Tibetan Plateau Research, Chinese Academy of Sciences, Beijing, China. 11CAS Center for Excellence in Tibetan Plateau Earth Sciences, Beijing, China. 12Department of F.A. Forel for Environmental and Aquatic Sciences, University of Geneva, Geneva, Switzerland. 13These authors contributed equally: Guoxiong Zheng, Simon Keith Allen. ✉e-mail: [email protected]; [email protected]

Nature Climate Change | VOL 11 | May 2021 | 411–417 | www.nature.com/natureclimatechange 411 Articles NaTurE CLImaTE CHanGE

Longitude 70° E 80° E 90° E 100° E 110° E

China

Ili East Tien Shan India ya ar yr D West Tien Shan 40° N S Gobi Interior im Tar w Hissar Alay o l l e A Y mu Qilian Shan D r ya a Pamir

East Kun Lun

Latitude West Kun Lun Hindu Kush Karakoram e us tz d Ya ng In

South and East Tibet Glacial lake West Himalaya Inner Tibet Sawlee GTN-G region n 30° N Tibetan Plateau Aggregated lake area Interior tr a Moraine Brahmapu dammed 250 km2 Central Himalaya Hengduan Shan M

y e d k d o East Himalaya a n G w g

CP8.5 a

a s r

Ice free e Present RCP2.6 RCP4.5 n I R g

Fig. 1 | Region-wide present and projected glacial lakes to 2050, 2100 and on an ice-free Third Pole. Map showing the geographical extent of the Third Pole and the spatial distribution of its present glacial lakes. Bar charts in different colours denote the present and potential future glacial lake areas (present plus projected results under three RCPs, and under the ice-free scenario) that were aggregated into the Global Terrestrial Network for Glaciers (GTN-G) regions46, respectively. The proportional area of present moraine-dammed glacial lakes to all present glacial lakes is shown in grey. Dashed lines show estimated changes in 2050.

Shan mountains and ranges of inner and southeast Tibet as well as Historical GLOF activity the Tien Shan mountains. The Hengduan Shan holds the largest As GLOF events across the Third Pole have been heavily reported number of glacial lakes (6,571; 436.2 ± 0.5 km2), whereas the east- in previous studies18–20,29, we systematically collected and collated ern Himalaya has the most moraine-dammed glacial lakes (1,533; these records to better understand their spatial distributions and 183.7 ± 0.4 km2). At the basin scale, glacial lakes are most abundant the characteristics of different source lakes, which were categorized in the Brahmaputra River Basin (9,665; 778.8 ± 0.8 km2), followed into moraine, ice, bedrock and complex ones depending on the dam by the Indus, Yangtze, Ganges, Amu Darya and Salween river basins type (Fig. 2a; see Supplementary Table 3 for all records assembled). (all with >1,000 lakes, Supplementary Fig. 3 and Supplementary On the basis of this dataset, we found that at least 296 GLOF events Table 2). The Brahmaputra River Basin is also home to the largest from 109 glacial lakes occurred over the Third Pole since 1560 number of moraine-dammed glacial lakes (2,331; 212.0 ± 0.4 km2). ce. GLOFs in the eastern Himalaya are primarily from the fail- Glacial lakes on the Third Pole have undergone a notable ures of moraine-dammed glacial lakes (Fig. 2c and Supplementary increase in total area and number since the 1990s5–7. Our anal- Table 3) and account for ~45% of the past GLOF sources alone. yses show an overall increase of 6.8 ± 0.1% in glacial lake area GLOFs resulting from ice-dammed glacial lakes are all reported and 5.9% in lake number across the Third Pole between 1990 in the Pamir–Karakoram regions (Fig. 2a), where the prevalent and 2015 (Supplementary Fig. 4 and Supplementary Table 1). glacier surges have led to repetitive blockage and eventual release We propose that this recent increase has principally been driven of glacial lakes19. Although most past GLOF sources were related by an expansion of moraine-dammed glacial lakes that char- to moraine-dammed glacial lakes (74 out of 109, Fig. 2b), ice- or actize many of the long, flat, debris-covered glacier tongues in complex-dammed flood sources (34 out of 109, Fig. 2b) contributed the region. To test this hypothesis, we split the glacial lakes into to more than half of the past outbursts (187 out of 296, Fig. 2b) due two categories: moraine-dammed and other-dammed glacial to the repetitive and high-frequency nature of these events. This is lakes (Methods). We find that if only the moraine-dammed gla- exemplified in the case of Kyagar Tsho (see Fig. 2a for its location cial lakes are considered, the area gain is substantially larger at and Supplementary Table 3), situated in the headwaters of Yarkant 31.3 ± 0.3%, with an increase in number of 26.7%. In contrast, the River, which produced at least 33 outburst floods in the past cen- size of other-dammed glacial lakes remains virtually unchanged tury with an accelerating trend during recent decades32. Practical (‒0.02 ± 0.2%), with a slight decrease in number of ‒1.4%. Still, experience has proven that real-time monitoring and early warning changes show clear heterogeneity both at regional and basin systems are effective in forecasting such active and high-frequency scales (Supplementary Figs. 4 and 5). floods resulting from ice dam breaches32,33. The complex-dammed

412 Nature Climate Change | VOL 11 | May 2021 | 411–417 | www.nature.com/natureclimatechange NaTurE CLImaTE CHanGE Articles

Longitude a 70° E 80° E 90° E 100° E 110° E

b GLOF frequency c 1 2 1 Kazakhstan 8 11 26 4 Uzbekistan 74 3 1 Tien Shan 112 26 7 96 108 Kyrgyzstan Merzbacher Lake 1 1 5 3 4 1 40° N 6 3 Shakhimardan 8 74 Tajikistan Catchment 113 Tarim 62 Basin Pamir 49 Karakoram GLOF source K2 Kunlun Shan Dam type GTN-G region Afghanistan 8,611 m Kyagar Tsho Moraine Hissar Alay South and East Tibet Hindu Kush Hindu Kush Latitude Ice Pamir Central Himalaya Karakoram Reported GLOFs Bedrock West Tien Shan East Himalaya Complex Inner Tibet Hengduan Shan Dam type Tibetan d Moraine Plateau 80 74 19 Ice All sources Bedrock Pakistan Disappeared Tibet/China 60 Complex after burst Hengduan Shan Transboundary GLOF threat 30° N Potentially dangerous glacial lake 40 Poiqu River Basin Present Future Himalaya Mount Everest 20 Modelled GLOF ﬂow path Nepal 8,848 m Number of sources 8 6 Cross-border point India 1 0 Cirenma Co 0 International border Moraine Bedrock Complex Disputed areas 500 km Bhutan Myanmar e 1 2 Gyirong Tibet/China Tingri Kangmar Jilong Nyalam Lhozhag 3 Comai Dinggye Gamba 4 Mount Everest 5 8,848 m Cona Latitude 28° N

Kathmandu Zhangmu Andhari Chomo Thimphu County/town Nepal India Bhutan 100 km

85° E 90° E Longitude

Fig. 2 | Reported historical GLOFs and potential transboundary threats on the Third Pole. a, Map showing the spatial distribution of recorded GLOF sources by lake dam type as well as present and projected (ice-free scenario) glacial lakes with possible transboundary GLOF threats across the Third Pole. For ice-dammed GLOF sources, only those with known geographic coordinates are shown (see Supplementary Table 3 for complete records). The black box near the bottom of the panel is the location of e. b, Double doughnut chart representing the number of past GLOF sources and flood frequency by lake dam type. c, Double doughnut chart showing the number of past GLOF sources and flood frequency per region. d, Statistics of GLOF sources by lake dam type. Ice-dammed cases were not counted owing to their repetitive nature. e, Amplified map showing a hotspot of potential transboundary GLOF threat between China and Nepal, and a historical GLOF hotspot in the eastern Himalaya. The circled numbers represent five concentrated regions with potential transboundary GLOF threats. The capital cities of Nepal and Bhutan are indicated with yellow squares. The left black box is the location of Supplementary Fig. 13 and the right black box is the location of Supplementary Fig. 14. Base maps: Google, Europa Technologies. glacial lakes are usually positioned on the surface of debris-covered the first quantitative assessment of potential GLOF hazard and risk glaciers and are held back by a mix of ice and moraine. These lakes for a total of 6,958 existing moraine-dammed glacial lakes across often vanish fully after an outburst (Fig. 2d), but some exceptions the Third Pole and provide hierarchical classifications. This study exist. For instance, Merzbacher Lake (see Fig. 2a for the specific compliments numerous national- to regional-scale GLOF haz- site and Supplementary Table 3), located upstream of Aksu River, ard and risk inventories that have previously been undertaken has produced at least 66 recorded outbursts over the last century (Supplementary Table 4), allowing a comparison of GLOF hazard because of its peculiar geographic situation and water supply and risk across the entire Third Pole. The model considers GLOF modes14,34. The occurrence of floods from such lakes is sudden and hazard on the basis of well-established conditioning and trigger- difficult to monitor and mitigate, as drainage can occur within or ing factors30 and identifies infrastructure and communities at risk underneath the glacier34. within downstream simulated flood paths. In contrast to previous work, we evaluate model reliability by validating classifica- Present hotspots of GLOF hazard and risk tion results against those glacial lakes that have generated outburst Based on a conceptual model of GLOF hazard and risk optimized floods in the past. The model successfully classified GLOF hazard as for large-scale automated application (Methods), we implemented high or very high for 46 out of the 48 existing glacial lakes (that is,

Nature Climate Change | VOL 11 | May 2021 | 411–417 | www.nature.com/natureclimatechange 413 Articles NaTurE CLImaTE CHanGE

a b Aggregated hazard value Aggregated risk value 500 250 250 125 100 50 50 25 <10 <5

GLOF hazard GLOF risk

VL VH VL VH L H L H M M c 400 Hazard class 1 1 9 37 Very low Low 300 Medium High Very high 200 Historical GLOFs (n = 48)

Frequency ( n = 6,958) 100

0 0 0.2 0.4 0.6 0.8 1.0

Normalized hazard value

Fig. 3 | Region-wide present GLOF hazard and risk across the Third Pole. a,b, Pie charts showing the proportion of different GLOF hazard (a) and risk (b) levels per region. VL, very low; L, low; M, medium; H, high; VH, very high. Aggregated hazard and risk values are the total sums of normalized values within each region. c, Histogram presenting the distribution of GLOF hazard levels across all moraine-dammed glacial lakes assessed. The vertical dashed lines indicate the distribution of assessed GLOF hazard values for glacial lakes where historical GLOFs have been recorded, which were further used to validate the overall model performance. The dots refer to the number of GLOF sources for a given hazard value.

95.8% accuracy, Fig. 3c and Supplementary Table 6), from which mapping of roads and other exposed elements over inner and south- the past GLOFs originated. For example, Jialong Co and Cirenma east Tibet is considered less complete than elsewhere35, these results Co, located in eastern Himalaya, have been the origin of several correspond well with other approaches based on gridded data GLOFs since 1960, and are both ranked within the 99th percen- including population density and cropland area that have likewise tile of the most hazardous lakes. Overall, our assessment indicates identified highest risk levels in central and eastern Himalaya36. At that one-third (2,323) of all glacial lakes assessed have high to very the basin scale, the Brahmaputra River Basin has the highest level in high hazard levels, and one in six (1,203) glacial lakes would pose terms of GLOF hazard, followed by Ganges, Indus, Salween, Ili, Amu a potentially high to very high risk to downstream communities. Darya river basins and Tibetan Plateau interior (Supplementary More than one-third (2,619) of glacial lakes assessed appear not to Fig. 7a). However, with respect to GLOF risk, the Ganges River Basin be dangerous as no downstream infrastructures are evident within has the highest level, followed by Brahmaputra, Indus, Amu Darya, the flood paths (see Supplementary Table 5 for all infrastructures Ili, Salween and Syr Darya river basins (Supplementary Fig. 7b). considered). On a regional scale, the eastern Himalaya has the highest value in terms of GLOF hazard, which is twice that of the adja- Anticipation of future changes cent Hengduan Shan, inner Tibet, central and western Himalaya, We expand our assessment to consider how and where GLOF haz- and southeast Tibet; three to four times that of the Tien Shan ard and risk will evolve under continued global warming and glacier areas; and seven to eight times that of the Pamir, Hindu Kush and shrinking. Future glacial lake formation and associated hazard and

Karakoram (Fig. 3a). For GLOF risk, the eastern Himalaya likewise risk are considered under three different CO2 emission scenarios has the highest value, being two to three times that of the central (Representative Concentration Pathways, RCP2.6, RCP4.5 and Himalaya, Hengduan Shan and western Himalaya; four times that RCP8.5)37 and projections of a glacier model38 for the middle (2050) of southeast Tibet and western Tien Shan; and six to eight times that and end (2100) of the twenty-first century (Methods). These results of inner Tibet, Hindu Kush, eastern Tien Shan and Pamir (Fig. 3b). are compared with an ice-free scenario which assumes that the Notably, some regions such as inner and southeast Tibet present Third Pole’s glaciers have totally vanished. Our modelling suggests high levels of GLOF hazard (Fig. 3a), but lower levels of GLOF that >13,000 new glacial lakes with a combined maximum area of risk (Fig. 3b) owing to much lower exposure values compared with ~1,510 km2 and a combined maximum volume of ~50 km3 could the central and eastern Himalaya (Supplementary Fig. 6). While potentially emerge in an ice-free environment (Supplementary

414 Nature Climate Change | VOL 11 | May 2021 | 411–417 | www.nature.com/natureclimatechange NaTurE CLImaTE CHanGE Articles

Table 1). This includes the expansion of existing glacial lakes to scenario. While lake volumes will on average be larger than today, reach their topographically constrained extent, and potential new it is clear that even in the future, it is not necessarily the largest glacial lakes forming in depressions within the former bed of gla- lakes that will be the biggest threat to communities (Supplementary ciers. In total, ~47% of this additional lake area would have already Fig. 10). At the basin scale, the future increase in GLOF risk will emerged by 2050, and 86% by 2100 under RCP8.5 (or 44% and 56%, be greatest in the Indus, Ganges and Amu Darya river basins, with respectively, under RCP2.6) (Supplementary Table 1). This future the Ganges River Basin maintaining the highest risk levels under glacial lake development is mostly clustered along the Pamir– all future scenarios (Supplementary Figs. 11 and 12). By 2100, the Karakoram–Kunlun Shan–Himalaya ranges as well as the western Indus River Basin will exhibit the highest GLOF hazard levels (cur- Tien Shan mountains (Fig. 1 and Supplementary Table 1), where the rently exhibited by the Brahmaputra River Basin), with close to a existing glaciers contribute ~74% of the total Third Pole glacier area fourfold increase from today under the RCP4.5 scenario. (Supplementary Table 1). There is notable regional variation in projected glacial lake development from east to west, with regions such Potential transboundary threats as southeast Tibet and eastern Himalaya already revealing close to The mountain ranges of the Tibetan Plateau and surrounding areas their maximum lake area by 2050, under all RCP scenarios (Fig. 1). span 11 nations (Fig. 2a), giving rise to potential transboundary nat- In contrast, in Karakoram, Pamir and western Himalaya, lake area ural disasters, especially GLOFs. In fact, several relevant cases have will continue to increase strikingly into the late twenty-first cen- been reported, for instance, in the transboundary Poiqu River Basin tury and beyond. On a basin scale, the future glacial lakes will be between China and Nepal22 and the Shakhimardan catchment between expected chiefly at the headwaters of the Indus and Tarim river Kyrgyzstan and Uzbekistan42 (Fig. 2a). A well-known outburst from basins, followed by Amu Darya, Ganges, Brahmaputra river basins Cirenma Co (Fig. 2a) in 1981 caused serious devastation, includ- and the interior ranges of the Tibetan Plateau (~89% of all projected ing but not limited to the destruction of the Sino–Nepal Friendship glacial lakes, Supplementary Fig. 3 and Supplementary Table 2). The bridge and part of the Nepal Sun Koshi hydropower station, and the Brahmaputra, Ganges and Yangtze river basins almost approach loss of 200 lives in Nepal43. Recognizing the broader transboundary their maximum lake area in 2050 under the three CO2 emission sce- threat across the Third Pole, we identified where simulated flood flow narios, whereas lakes in the Indus and Tarim river basins as well as paths from existing moraine-dammed glacial lakes intersected with the Tibetan Plateau interior will continue to expand beyond 2100. national borders. Consequently, a total of 438 glacial lakes were found To assess the possible implications of this future glacial lake to have the potential to produce a transboundary flood, of which 191 development on GLOF hazard and risk, we applied the same model were classified to be in the high or very high risk classes (Fig. 2a). as was used for the present glacial lakes, with some slight modifi- Further analysis revealed that ~86% of these potentially dangerous cations to account for future unknowns (Methods). For the whole lakes are located between China and Nepal, forming five hotspots Third Pole region, results indicate that future GLOF hazard and risk of transboundary floods (Fig. 2e). It can be inferred from the illus- will increase by almost threefold relative to current conditions, with trated example of one of these hotspots shown in Supplementary Fig. obvious regional variations (Fig. 4 and Supplementary Fig. 8). This 13 that a possible outburst will have serious effects on downstream is considered a conservative maximum estimate given that glacial transboundary settlements and infrastructures. lakes will appear gradually, and an unknown quantity of moraine The transboundary GLOF threat will amplify in the future. dams could breach, and lakes thereby drain, before other lakes have Analysis suggests that the number of future potential transbound- emerged. Nonetheless, under the high-emission scenario RCP8.5, ary flood sources under the ice-free scenario will roughly double much of the Third Pole could already be approaching a state of ‘peak (an additional 464 lakes), with 211 of these lakes classified in the risk’ by the end of the twenty-first century, or even mid-century in high or very high risk classes (Fig. 2a). The border region between some regions, with risk levels comparable to even the ice-free sce- China and Nepal will remain a major hotspot (~42% of all these nario (Fig. 4). In an ice-free environment, average glacial lake vol- future dangerous glacial lakes), while the border region between umes will be ~80% larger than today, while topographic potential for Tajikistan and Afghanistan emerges clearly as an additional poten- ice/rock avalanche triggering will increase by ~230%, as lakes will tial transboundary hotspot (~36%, up from ~5% at present). be positioned closer to steep mountain headwalls (Supplementary Fig. 9). This suggests that the increase in potential GLOF frequency Summary and implications will generally be more important than the future change in magni- We conclude that while the most dangerous hotspots of GLOF hazard tude. Meanwhile, in terms of risk, although overall exposure levels and risk are now located within the eastern Himalaya, consistent with will increase due to the larger number of lakes, future lakes will on previous studies21,23,36,44 and recorded GLOF events (Supplementary average have lower exposure levels (‒40% compared with today) Fig. 10; see Supplementary Fig. 14 for a visual example in the east- as they are located higher, and farther away from existing commu- ern Himalaya), over the course of the twenty-first century, addi- nities. As a consequence, average risk values per lake will be 14% tional hotspots will emerge in the regions to the west, particularly lower in the future, while the overall number of lakes classified to under high-emission scenarios. For Karakoram and Pamir, where have high or very high risk levels will increase from 1,203 to 2,963 glaciers have recently exhibited a stable or slightly positive mass bal- lakes (that is, by a factor of 2.5). This analysis does not consider ance45, repetitive, high-frequency outburst events associated with future changes in the distribution of population and infrastructure, ice-dammed glacial lakes will probably remain the dominant GLOF particularly hydropower, that is expanding higher into alpine val- threat for several decades to come. However, the eventual depletion leys39. Likewise, tourism expansion into remote mountain valleys of glaciers in these regions projected during the twenty-first century could further enhance risks, while outmigration from rural areas will see conditions becoming less conducive for glacier surging and could reduce risk in some cases40,41. Collectively, the Karakoram, ice-dam formation19, and the threat of moraine- or bedrock-dammed Pamir and western Himalaya will be the regions with the most sub- glacial lakes substantially increasing. It is also important to emphasize stantial increase in GLOF hazard, while in terms of risk, the larg- that climate and related glacial changes as assessed here are just one est increases will be in central Himalaya, Karakoram and western driver of future GLOF risk; in some regions, future changes in expo- Himalaya (Fig. 4 and Supplementary Fig. 8). The eastern Himalaya sure and societal vulnerability could potentially lead to different risk will remain the primary GLOF hotspot under all assessed future scenarios, especially where political, economic and social conditions scenarios, while the emergence of the Karakoram as an additional facilitate or impede effective disaster risk reduction. Ultimately, deter- major hotspot of GLOF hazard and risk is most notable over lon- mining exactly which lakes are the highest priority for response actions ger timescales under RCP4.5 and RCP8.5, and under the ice-free should be guided by consensus across multiple lines of evidence,

Nature Climate Change | VOL 11 | May 2021 | 411–417 | www.nature.com/natureclimatechange 415 Articles NaTurE CLImaTE CHanGE

Longitude 70° E 80° E 90° E 100° E 110° E

Total risk value 3,861 3,597 3,197 2,705 1,510 Ili

East Tien Shan Third Pole ya ar D 40° N Syr West Tien Shan im Tar w o l Hissar Alay l e A Y mu Qilian Shan D a r ya Pamir

East Kun Lun

Latitude Karakoram West Kun Lun Hindu Kush e s tz du Ya ng In South and East Tibet

Inner Tibet West Himalaya Sawlee 30° N n Aggregated risk value r a Ice free ut Brahmap Hengduan Shan RCP8.5 RCP4.5 Central Himalaya

RCP2.6 M y e d k Present 600 d o East Himalaya a n G w g a

s r

a e r ng I

Fig. 4 | Region-wide future changes in GLOF risk to 2050, 2100 and on an ice-free Third Pole. Bar plots indicate projected changes in GLOF risk per region from the present to 2050 and 2100 (under three RCPs) as well as under the ice-free scenario. Dashed lines show estimated changes to 2050. The future GLOF risk was estimated based on currently known infrastructures that are exposed to modelled flood flow paths from potential future lakes and includes the assessed risk from present moraine-dammed glacial lakes (that is, these lakes are assumed to remain in the future). Inset: changes in GLOF risk over the whole Third Pole; note that the scale differs from the regional risk values. considering not only the wealth of information on existing GLOF 2. Bolch, T. et al. Te state and fate of Himalayan glaciers. Science 336, hazard and risk (Supplementary Table 4), but also future emerging 310–314 (2012). 3. Sorg, A., Bolch, T., Stofel, M., Solomina, O. & Beniston, M. Climate change threats. Particularly where existing and emerging transboundary risks impacts on glaciers and runof in Tien Shan (Central Asia). Nat. Clim. have been identified, risk reduction strategies will require high lev- Change 2, 725–731 (2012). els of international planning and cooperation. These transboundary 4. Hock, R. et al. in IPCC Special Report on the Ocean and Cryosphere in a regions represent vital economic corridors of activity, but some suffer Changing Climate (eds Portner H.-O. et al.) Ch. 2 (IPCC, 2019). from political tensions and challenges that can negatively affect timely 5. Zhang, G., Yao, T., Xie, H., Wang, W. & Yang, W. An inventory of glacial lakes in the Tird Pole region and their changes in response to global data sharing, communication and coordination of early warning and warming. Glob. Planet. Change 131, 148–157 (2015). disaster preparedness. We appeal to the relevant nations and interna- 6. Nie, Y. et al. A regional-scale assessment of Himalayan glacial lake changes tional research communities to work together to prevent and mitigate using satellite observations from 1990 to 2015. Remote Sens. Environ. 189, the damages and losses that GLOFs bring to the Third Pole region, 1–13 (2017). and to remain flexible and alert to the new challenges that could 7. Shugar, D. H. et al. Rapid worldwide growth of glacial lakes since 1990. Nat. Clim. Change 10, 939–945 (2020). emerge under a future warmer climate. 8. Linsbauer, A. et al. Modelling glacier-bed overdeepenings and possible future lakes for the glaciers in the Himalaya—Karakoram region. Ann. Glaciol. 57, Online content 119–130 (2016). Any methods, additional references, Nature Research report- 9. Haeberli, W. et al. New lakes in deglaciating high-mountain ing summaries, source data, extended data, supplementary infor- regions — opportunities and risks. Climatic Change 139, 201–214 (2016). mation, acknowledgements, peer review information; details of 10. Farinotti, D., Round, V., Huss, M., Compagno, L. & Zekollari, H. Large author contributions and competing interests; and statements of hydropower and water-storage potential in future glacier-free basins. Nature data and code availability are available at https://doi.org/10.1038/ 575, 341–344 (2019). s41558-021-01028-3. 11. Begam, S., Sen, D. & Dey, S. Moraine dam breach and glacial lake outburst food generation by physical and numerical models. J. Hydrol. 563, Received: 2 June 2020; Accepted: 16 March 2021; 694–710 (2018). 12. Cook, K. L., Andermann, C., Gimbert, F., Adhikari, B. R. & Hovius, N. Published online: 6 May 2021 Glacial lake outburst foods as drivers of fuvial erosion in the Himalaya. Science 362, 53–57 (2018). References 13. Dubey, S. & Goyal, M. K. Glacial lake outburst food hazard, downstream 1. Yao, T. et al. Diferent glacier status with atmospheric circulations in Tibetan impact, and risk over the Indian Himalayas. Water Resour. Res. 56, Plateau and surroundings. Nat. Clim. Change 2, 663–667 (2012). e2019WR026533 (2020).

416 Nature Climate Change | VOL 11 | May 2021 | 411–417 | www.nature.com/natureclimatechange NaTurE CLImaTE CHanGE Articles

14. Carrivick, J. L. & Tweed, F. S. A global assessment of the societal impacts of 31. Haeberli, W., Schaub, Y. & Huggel, C. Increasing risks related to landslides glacier outburst foods. Glob. Planet. Change 144, 1–16 (2016). from degrading permafrost into new lakes in de-glaciating mountain ranges. 15. Emmer, A. GLOFs in the WOS: bibliometrics, geographies and global trends Geomorphology 293, 405–417 (2017). of research on glacial lake outburst foods (Web of Science, 1979–2016). 32. Yin, B., Zeng, J., Zhang, Y., Huai, B. & Wang, Y. Recent Kyagar glacier lake Nat. Hazards Earth Syst. Sci. 18, 813–827 (2018). outburst food frequency in Chinese Karakoram unprecedented over the last 16. Yang, K. et al. Recent climate changes over the Tibetan Plateau and their two centuries. Nat. Hazards 95, 877–881 (2019). impacts on energy and water cycle: a review. Glob. Planet. Change 112, 33. Haemmig, C. et al. Hazard assessment of glacial lake outburst foods from 79–91 (2014). Kyagar glacier, Karakoram Mountains, China. Ann. Glaciol. 55, 34–44 (2014). 17. Harrison, S. et al. Climate change and the global pattern of moraine-dammed 34. Shangguan, D. et al. Quick release of internal water storage in a glacier leads glacial lake outburst foods. Cryosphere 12, 1195–1209 (2018). to underestimation of the hazard potential of glacial lake outburst foods 18. Hewitt, K. & Liu, J. Ice-dammed lakes and outburst foods, Karakoram from Lake Merzbacher in Central Tian Shan Mountains. Geophys. Res. Lett. Himalaya: historical perspectives on emerging threats. Phys. Geogr. 31, 44, 9786–9795 (2017). 528–551 (2010). 35. Barrington-Leigh, C. & Millard-Ball, A. Te world’s user-generated road map 19. Bhambri, R. et al. Ice-dams, outburst foods, and movement heterogeneity of is more than 80% complete. PLoS ONE 12, e0180698 (2017). glaciers, Karakoram. Glob. Planet. Change 180, 100–116 (2019). 36. Wang, S., Qin, D. & Xiao, C. Moraine-dammed lake distribution and outburst 20. Nie, Y. et al. An inventory of historical glacial lake outburst foods in the food risk in the Chinese Himalaya. J. Glaciol. 61, 115–126 (2015). Himalayas based on remote sensing observations and geomorphological 37. Meinshausen, M. et al. Te RCP greenhouse gas concentrations and their analysis. Geomorphology 308, 91–106 (2018). extensions from 1765 to 2300. Climatic Change 109, 213 (2011). 21. Allen, S. K., Zhang, G., Wang, W., Yao, T. & Bolch, T. Potentially dangerous 38. Huss, M. & Hock, R. Global-scale hydrological response to future glacier glacial lakes across the Tibetan Plateau revealed using a large-scale automated mass loss. Nat. Clim. Change 8, 135–140 (2018). assessment approach. Sci. Bull. 64, 435–445 (2019). 39. Schwanghart, W., Worni, R., Huggel, C., Stofel, M. & Korup, O. Uncertainty 22. Khanal, N. R. et al. A comprehensive approach and methods for glacial lake in the Himalayan energy–water nexus: estimating regional exposure to glacial outburst food risk assessment, with examples from Nepal and the lake outburst foods. Environ. Res. Lett. 11, 074005 (2016). transboundary area. Int. J. Water Resour. Dev. 31, 219–237 (2015). 40. Ziegler, A. D. et al. Pilgrims, progress, and the political economy of disaster 23. Veh, G., Korup, O. & Walz, A. Hazard from Himalayan glacier lake outburst preparedness — the example of the 2013 Uttarakhand food and Kedarnath foods. Proc. Natl Acad. Sci. USA 117, 907–912 (2020). disaster. Hydrol. Process. 28, 5985–5990 (2014). 24. Fujita, K. et al. Potential food volume of Himalayan glacial lakes. 41. von Dach, W. et al. Safer Lives and Livelihoods in Mountains: Making the Nat. Hazards Earth Syst. Sci. 13, 1827–1839 (2013). Sendai Framework for Disaster Risk Reduction Work for Sustainable Mountain 25. Neupane, R., Chen, H. & Cao, C. Review of moraine dam failure mechanism. Development (Centre for Development and Environment (CDE), University of Geomat. Nat. Hazards Risk 10, 1948–1966 (2019). Bern, with Bern Open Publishing (BOP), 2017). 26. Allen, S. K., Rastner, P., Arora, M., Huggel, C. & Stofel, M. Lake outburst 42. Petrakov, D. A. et al. Putting the poorly documented 1998 GLOF disaster in and debris fow disaster at Kedarnath, June 2013: hydrometeorological Shakhimardan River valley (Alay range, Kyrgyzstan/Uzbekistan) into triggering and topographic predisposition. Landslides 13, 1479–1491 (2016). perspective. Sci. Total Environ. 724, 138287 (2020). 27. Zaginaev, V. et al. Reconstruction of glacial lake outburst foods in northern 43. Wang, W. et al. Integrated hazard assessment of Cirenmaco glacial Tien Shan: implications for hazard assessment. Geomorphology 269, lake in Zhangzangbo Valley, Central Himalayas. Geomorphology 306, 75–84 (2016). 292–305 (2018). 28. Emmer, A. & Cochachin, A. Te causes and mechanisms of 44. Wang, S., Che, Y. & Xinggang, M. Integrated risk assessment of glacier lake moraine-dammed lake failures in the Cordillera Blanca, North American outburst food (GLOF) disaster over the Qinghai–Tibetan Plateau (QTP). Cordillera, and Himalayas. AUC Geographica 48, 5–15 (2013). Landslides 17, 2849–2863 (2020). 29. Veh, G., Korup, O., von Specht, S., Roessner, S. & Walz, A. Unchanged 45. Azam, M. F. et al. Review of the status and mass changes of Himalayan– frequency of moraine-dammed glacial lake outburst foods in the Himalaya. Karakoram glaciers. J. Glaciol. 64, 61–74 (2018). Nat. Clim. Change 9, 379–383 (2019). 46. GTN-G Glacier Regions (Global Terrestrial Network for Glaciers, 2017); 30. GAPHAZ Assessment of Glacier and Permafrost Hazards in Mountain Regions https://doi.org/10.5904/gtng-glacreg-2017-07 — Technical Guidance Document (Standing Group on Glacier and Permafrost Hazards in Mountains (GAPHAZ) of the International Association of Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in Cryospheric Sciences (IACS) and the International Permafrost Association published maps and institutional affiliations. (IPA), 2017). © The Author(s), under exclusive licence to Springer Nature Limited 2021

Nature Climate Change | VOL 11 | May 2021 | 411–417 | www.nature.com/natureclimatechange 417 Articles NaTurE CLImaTE CHanGE

Methods overdeepenings-filled DEM and the original glacier-bed topography. The mean Glacial lake delineation and lake changes. We used Landsat satellite imagery and maximum depth of each glacial lake was also estimated using the detected to outline the boundaries of glacial lakes that are defned as water bodies formed depression extent and its bathymetry raster. In general, ice-thickness models 59 primarily by glaciation and fed by glacial/snow melts, and are situated on the have proven robust in detecting the locations of overdeepenings , although local surface, adjacent or downstream of a glacier, or on palaeoglaciation landforms47 uncertainties in ice-thickness estimates (and thereby volume of the overdeepenings 60 across the Tird Pole. Two time windows, 1990 ± 4 and 2015 ± 1, were chosen to and associated future lake) could be on the order of ±100% for individual cases . reveal the present state of glacial lakes and their changes over the last ~25 years. Likewise, dam composition (moraine or bedrock) cannot be distinguished for Te selection was determined largely by the availability of Landsat datasets in such future lakes, and lake volumes will be ultimately determined by dam geometry and a large area. To minimize the efects of ice or seasonal snow cover on glacial lake the presence of an outlet channel. mapping in these high elevations, we picked images mostly from the warm season To estimate the time horizon in which potential future glacial lakes could (June to November), which corresponds roughly to the highest and most stable emerge, we simulated glacier recession based on an ensemble of 14 global 61 water areas/levels of the lakes in a hydrological year48. Overall, we used a total of circulation models of the Climate Model Intercomparison Project Phase 5 (ref. ) 403 Landsat 4–5 Tematic Mapper (TM) scenes (97.8% from the warm season) for under three different CO2 emission scenarios — RCP2.6, RCP4.5 and RCP8.5, in the 1990 epoch and 294 Landsat 8 Operational Land Imager (OLI) scenes (100% addition to the hypothetical end case with a completely ice-free Third Pole. Two in the warm season) for the 2015 period (Supplementary Figs. 1 and 2a,b). We also major time horizons over the course of the twenty-first century, 2050 and 2100, 38 gave priority to images with no or little cloud coverage, that is, an average cloud were considered. The Global Glacier Evolution Model (GloGEM) was employed cover percentage of 7.7% and 4.2% for 1990 and 2015, respectively (Supplementary to model the evolution of all individual RGI 6.0 glaciers based on projected Figs. 1 and 2c). Landsat images were obtained from the US Geological Survey climate forcing until 2100. GloGEM resolves all components of the glacier surface (USGS) data portal with a spatial resolution of 30 m. Images were frst converted mass balance and computes changes in glacier geometry (thickness, length) in from the digital numbers to the top of atmosphere. Landsat 8 OLI images annual time steps. Results for the ensemble of global circulation models have were further fused using a nearest neighbour difusion-based pan-sharpening been averaged for the three considered RCPs. The minimum elevation for each algorithm49 to improve their spatial resolution to 15 m. glacier at both 2050 and 2100 under each RCP scenario was calculated and used An automated water body classification algorithm50 based on global–local to determine the exposed overdeepenings at each time step. We assume that the threshold segmentation for water indices greyscale images was applied to extract potential glacial lake is forming or has formed if the modelled minimum elevation all water bodies from the Landsat images. A slope threshold of <20° as well as of its parent glacier is greater than or equal to the elevation of the modelled a shaded relief threshold of >0.25 were employed to alleviate the disturbances overdeepening that was retrieved based on the filled glacier-bed topographies. from mountain shadows50. Subsequently, careful visual inspection and manual re-editing were carried out to remove other water bodies (for example, reservoirs, GLOF hazard assessment and validation. Moraine-dammed glacial lakes 17 rivers and small streams) and to correct the wrongly mapped glacial lakes based on dominate past GLOFs recorded in most of the world , especially in the 20,29 corresponding Landsat scenes, online maps and other glacial lake datasets5,47,51–53 Himalaya , and their evolution can be directly linked to atmospheric warming 17 available for the Third Pole. A 10 km buffer5,52 around the Randolph Glacier and glacial recession . Hence, for the current GLOF hazard and risk assessment, 21 Inventory (RGI 6.0)54 was used to preliminarily determine the distribution of what we focused on such lakes. A conceptual model was used to assess hazard and could be considered glacial lakes. However, to ensure the integrity and reliability risk for 6,958 existing moraine-dammed proglacial lakes across the Third Pole. of our inventory, we also included additional glacial lakes in areas where glaciers The model does not apply to some glacial lakes situated on the surface or flanks of are missing from the RGI 6.0 dataset55. Both the glacial lake datasets for 1990 debris-covered glaciers, as predisposing and triggering factors in these cases are and 2015 were then cross-checked and attributed. The area of all glacial lakes different. Furthermore, we recognize the transition and complexity that occurs was calculated based on the universal transverse mercator (UTM) projected between different lake types, meaning that some degree of subjectivity in lake 21 coordinate system and their uncertainty (δ) was estimated using the formula56: classification cannot be avoided. The model defines three indices : the GLOF δ = P/G × G2/2 × 0.6872, where P is the perimeter of the glacial lake and G is the risk index as a function of the hazard index (combining likelihood and potential spatial resolution of the images used. Lastly, we selected all glacial lakes ≥0.01 km2 magnitude of a GLOF) and exposure index (potential for people and infrastructure and superimposed them in a high-resolution Google Earth 3D view mode together to be affected). Here we improved the model for large-scale implementation so that with the corresponding Landsat scenes to distinguish the moraine-dammed glacial it can run on a single glacial lake object iteratively. The large-scale assessment was lakes separately. The remaining lakes were grouped and termed ‘other-dammed based on a single UTM grid zone. All calculations and analyses described below 62 glacial lakes’. Lakes dammed with ice or lying on the surface of glaciers were not were based on the 90 m Multi-Error-Removed Improved-Terrain (MERIT) DEM , 63 included in the temporal change analyses due to their transient character and large which is proven to have the best performance on the Third Pole . 21 seasonal fluctuations. The potential hazard of a GLOF was defined based on four key factors , which 30 The relative change in glacial lake area between 1990 and 2015 and its encompass well-established processes that drive GLOF triggering and magnitude . uncertainty were estimated using the following equations: (1) Lake volume was calculated using a lake depth–area–volume empirical relationship64 and used as a proxy for possible maximum flood magnitude. (2) − = A2 A1 × % Total watershed area upstream of each glacial lake was used as an indicator of R A1 100 the potential for heavy rain and glacier/snow melt runoff to flow into the lake or 2 2 20,26 δA δA1 × cause it to overflow . (3) The likelihood of ice and/or rock avalanches triggering δR = − + R √( A2 A1 ) ( A1 ) an outburst was calculated based on the concept of topographic potential which includes: the potential for ice and/or rock to detach (parameterized by slope where R is the relative change in glacial lake area and δR is its uncertainty; A1, A2 angle); and the potential for the ice and/or rock avalanche to reach a glacial lake are the lake areas at the beginning and end of the period, respectively; and δA1, δA2 (parameterized by the overall trajectory slope or angle of reach). Due to the quality = 2 + 2 are their uncertainties,δA δA1 δA2 . of available glacier datasets at such a large scale, we did not distinguish whether the √ surrounding slopes were rock or ice-covered. Thus, we assumed that an avalanche Modelling of future glacial lakes. Glacier-bed topography, that is, ice-free was possible from any slope >30°, and the resulting avalanche could feasibly digital elevation model (DEM) can be estimated using model approaches that impact the lake where the overall slope trajectory is >14° (tangent angle = 0.25)65. determine local ice thickness based on the characteristics of surface topography (4) Downstream slope of the lake dam was considered to be an important factor and considerations of ice-flow dynamics8. The potential sites of future glacial controlling dam stability and the potential for self-destruction. It is also relevant lake formation under scenarios of complete deglaciation can then be projected by for the erosional capacity of a GLOF event in downstream areas. This factor detecting overdeepenings on the glacier-bed. Here we used the latest global glacier was calculated as the average horizontal downstream slope of each lake in three ice thickness dataset57 consisting of a consensus of five different ice thickness different buffer zones depending on different lake areas (A) intervals (A ≥ 0.1 km2, models as well as the 30 m Advanced Land Observing Satellite Global Digital 900 m; 0.1 > A ≥ 0.05 km2, 600 m; 0.05 > A ≥ 0.01 km2, 300 m). These intervals were Surface Model (ALOS AW3D30 v2.2) to extract the bed topography of all Third derived empirically based on careful analyses of a large number of lakes using Pole glaciers. Three analytical steps were performed to determine the possible sites high-resolution satellite images. of future glacial lake formation: (1) glacier areas with slopes <20° were mapped All factors were then normalized to 0 to 1 using the percent ranking method first as potential locations of glacier-bed overdeepenings as lakes normally form and assumed to have equal weight to the hazard of a GLOF event. For display and in flat topography; (2) depressions in the glacier bed were detected by filling comparative purposes, hazards of all glacial lakes were eventually classified into them using a classic filling algorithm58, while a slope raster was generated based five levels (very low, low, medium, high, very high) using the natural breaks (Jenks) on the filled glacier-bed topographies; and (3) depressions with a slope <1° were method and then were aggregated to each region and river basin based on the sum identified8 and intersected with the layer of possible overdeepenings obtained in of all lake hazard levels. the first step to generate the final layer of the possible future lakes. Ultimately, The verification of model outcomes is highly dependent on empirical we selected all modelled overdeepenings ≥0.01 km2 as the potential sites of future information21. Here, to verify the results from the GLOF hazard assessment and lake formation, to ensure comparability to the glacial lake inventories in 1990 and provide the historical context of GLOF activity, we compiled the most complete 2015. The area and volume of all modelled future glacial lakes were estimated inventory of past GLOFs over the Third Pole. Thus, we collected and re-evaluated based on a bathymetry raster that was derived from the difference between the almost all of the historical GLOF events (Supplementary Table 3) that have been

Nature Climate Change | www.nature.com/natureclimatechange NaTurE CLImaTE CHanGE Articles reported on the Third Pole and found a total of 296 GLOF events from 109 glacial 51. Wang, X. et al. Changes of glaciers and glacial lakes implying corridor-barrier lakes. Finally, 48 moraine-dammed glacial lakes that did not disappear after efects and climate change in the Hengduan Shan, southeastern Tibetan previous GLOFs and remained ≥0.01 km2 in the 2015 epoch were used to validate Plateau. J. Glaciol. 63, 535–542 (2017). the hazard assessment. 52. Wang, X. et al. Changes of glacial lakes and implications in Tian Shan, central Asia, based on remote sensing data from 1990 to 2010. Environ. Res. Lett. 8, GLOF exposure and risk. Exposure represents the presence of people, 044052 (2013). livestock, infrastructure and other assets that could be threatened by potential 53. Wang, X., Liu, Q., Liu, S., Wei, J. & Jiang, Z. Heterogeneity of glacial lake GLOF hazard21. Here we used infrastructures (such as buildings, highways, expansion and its contrasting signals with climate change in Tarim Basin, railways and historic places, see Supplementary Table 5 for all features used) to Central Asia. Environ. Earth Sci. 75, 696 (2016). characterize the communities and populations that might be affected within a 54. RGI Consortium Randolph Glacier Inventory — A Dataset of Global Glacier potential GLOF flow path, which was simulated using the GIS-based modified Outlines: Version 6.0: Technical Report (Global Land Ice Measurements from single-flow hydrological model66. All infrastructure features were downloaded Space (GLIMS), 2017); https://doi.org/10.7265/N5-RGI-60 from OpenStreetMap (accessed 1 August 2019) and were converted to a raster 55. Sakai, A. Brief communication: updated GAMDAM glacier inventory over grid. The maximum downstream flow distance for each lake path was determined high-mountain Asia. Cryosphere 13, 2043–2049 (2019). based on an empirically derived worst-case scenario defined by the angle of reach 56. Hanshaw, M. N. & Bookhagen, B. Glacial areas, lake areas, and snow lines from the source lake, with a tangent angle = 0.05 (3° angle of reach)67. Beyond this from 1975 to 2012: status of the Cordillera Vilcanota, including the flow distance, damages from a GLOF are not expected. The sum of the angle of Quelccaya Ice Cap, northern central Andes, Peru. Cryosphere 8, reach for each OpenStreetMap raster pixel that is exposed to the lake flow path 359–376 (2014). was aggregated as the quantified measure of exposure for each glacial lake. As 57. Farinotti, D. et al. A consensus estimate for the ice thickness distribution of of 2016, the OpenStreetMap was considered to be 83% globally complete and all glaciers on Earth. Nat. Geosci. 12, 168–173 (2019). increasing rapidly, although generally lower (yet relatively homogeneous) levels of 58. Planchon, O. & Darboux, F. A fast, simple and versatile algorithm to fll the completeness are seen over the Third Pole35. The exception is Nepal where intense depressions of digital elevation models. Catena 46, 159–176 (2002). humanitarian mapping efforts have led to near total completeness. 59. Linsbauer, A., Paul, F. & Haeberli, W. Modeling glacier thickness distribution The exposure values for each glacial lake were also normalized to 0 to 1 using and bed topography over entire mountain ranges with GlabTop: application the percent ranking method and then multiplied with the normalized hazard of a fast and robust approach. J. Geophys. Res. Earth Surf. 117, F03007 (2012). value to give the final risk level associated with each lake. The risk values of all 60. Farinotti, D. et al. How accurate are estimates of glacier ice thickness? Results glacial lakes (except those with an exposure value of zero indicating no risk to from ITMIX, the Ice Tickness Models Intercomparison eXperiment. downstream communities) were divided into five classes using the natural breaks Cryosphere 11, 949–970 (2017). (Jenks) method and aggregated to each region and river basin to obtain their 61. Taylor, K. E., Stoufer, R. J. & Meehl, G. A. An overview of CMIP5 and the overall risk levels. In the absence of reasonable proxy variables at such a large scale, experiment design. Bull. Am. Meteorol. Soc. 93, 485–498 (2012). the vulnerability component of risk was not considered here. 62. Yamazaki, D. et al. A high-accuracy map of global terrain elevations. Geophys. Res. Lett. 44, 5844–5853 (2017). Future GLOF hazard and risk assessments. The same modelling approach was 63. Liu, K. et al. Global open-access DEM performances in Earth’s most rugged also applied to assess future GLOF hazards and risks for all modelled glacial lakes region High Mountain Asia: a multi-level assessment. Geomorphology 338, formed at 2050 and 2100 under three different RCPs and the ice-free scenario. The 16–26 (2019). difference was that the current topography of the glacier areas was substituted by 64. Huggel, C., Kääb, A., Haeberli, W., Teysseire, P. & Paul, F. Remote sensing the modelled glacier-bed topography and then the selected factors were calculated based assessment of hazards from glacier lake outbursts: a case study in the for these future scenarios. However, because we cannot judge whether these Swiss Alps. Can. Geotech. J. 39, 316–330 (2002). potential glacial lakes will be dammed by moraine or bedrock, and resulting dam 65. Schneider, D., Huggel, C., Haeberli, W. & Kaitna, R. Unraveling driving geometries are highly uncertain, we used the average dam slope of present glacial factors for large rock–ice avalanche mobility. Earth Surf. Process. Landf. 36, lakes in each region or river basin to characterize all modelled lakes in the same 1948–1966 (2011). region or basin. Likewise, no robust basis exists to predict the future development 66. Huggel, C., Kääb, A., Haeberli, W. & Krummenacher, B. Regional-scale of downstream infrastructure and communities across all of the Third Pole; GIS-models for assessment of hazards from glacier lake outbursts: evaluation thus, the future risk assessment considers only the change in exposure of existing and application in the Swiss Alps. Nat. Hazards Earth Syst. Sci. 3, infrastructures and communities. 647–662 (2003). 67. Huggel, C., Haeberli, W., Kääb, A., Bieri, D. & Richardson, S. An assessment Data availability procedure for glacial hazards in the Swiss Alps. Can. Geotech. J. 41, The Landsat datasets are freely available from the USGS data portal (https://glovis. 1068–1083 (2004). usgs.gov/). The Randolph Glacier Inventory 6.0 (ref. 54) data are freely available at http://www.glims.org/RGI/. The ALOS Global Digital Surface Model (AW3D30 Acknowledgements v2.2) is freely available at https://www.eorc.jaxa.jp/ALOS/en/aw3d30/index.htm. We thank all those who made any data used here freely available. This work was The Multi-Error-Removed Improved-Terrain (MERIT) DEM is freely available funded by the Strategic Priority Research Program of the Chinese Academy of Sciences at http://hydro.iis.u-tokyo.ac.jp/~yamadai/MERIT_DEM/ on reasonable request. (XDA20030101) and the National Key Research and Development Program of China The OpenStreetMap data are freely obtained from http://www.openstreetmap.org/ (2017YFB0504204). The contribution of S.K.A. and G.Q.Z was partially supported by and available under the Open Database License (http://www.openstreetmap.org/ the Swiss National Science Foundation (IZLCZ2_169979/1), and counterpart grant copyright). The composite glacier ice thickness data are available at https://www. from the National Natural Science Foundation of China (21661132003). A special research-collection.ethz.ch/handle/20.500.11850/315707. The produced glacial acknowledgement goes to the China–Pakistan Joint Research Center on Earth Sciences lake inventories and modelled future lakes as well as relevant assessment results are that supported the implementation of this work. available at Zenodo under the identifier https://doi.org/10.5281/zenodo.4477945. Author contributions Code availability Conceptualization: G.Z., S.K.A., A.B., M.S. Data preparation and model calculations: The GLOF hazard and risk assessment models are available at Zenodo under the G.Z., A.B., M.H., G.Q.Z., J.L., Y.Y., T.Y., W.C. Funding acquisition: A.B., S.K.A., G.Q.Z. identifier https://doi.org/10.5281/zenodo.4477947. Additional model or code used Methodology: G.Z., S.K.A., M.H. Visualization: G.Z. Project administration: A.B., M.S. in this study are available from the corresponding authors on request. Writing, original draft: G.Z., S.K.A., J.B.-C., M.S. Writing, review, and editing: all authors.

References Competing interests 47. Maharjan, S. B. et al. Te Status of Glacial Lakes in the Hindu Kush Himalaya. The authors declare no competing interests. ICIMOD Research Report 2018/1 (International Centre for Integrated Mountain Development (ICIMOD), 2018). 48. Zhang, Y., Zhang, G. & Zhu, T. Seasonal cycles of lakes on the Tibetan Additional information Plateau detected by Sentinel-1 SAR data. Sci. Total Environ. 703, Supplementary information The online version contains supplementary material 135563 (2020). available at https://doi.org/10.1038/s41558-021-01028-3. 49. Sun, W., Chen, B. & Messinger, D. Nearest-neighbor difusion-based pan-sharpening algorithm for spectral images. Opt. Eng. 53, 013107 (2014). Correspondence and requests for materials should be addressed to A.B. or M.S. 50. Zhang, G. et al. Automated water classifcation in the Tibetan Plateau Peer review information Nature Climate Change thanks Rakesh Bhambri, Adam Emmer using Chinese GF-1 WFV data. Photogramm. Eng. Remote Sens. 83, and Kristyna Falatkova for their contribution to the peer review of this work. 509–519 (2017). Reprints and permissions information is available at www.nature.com/reprints.

Nature Climate Change | www.nature.com/natureclimatechange